cstr/hubert-large-ls960-ft-GGUF

HuBERT Large (GGUF)

GGUF conversion of facebook/hubert-large-ls960-ft for use with CrispASR.

Model Details

• Architecture: HuBERT — wav2vec2-style CNN (7L, 512-dim) + 24-layer transformer (1024-dim, 16 heads, pre-norm) + CTC head
• Parameters: ~316M
• Training: Self-supervised pre-training on LibriSpeech 960h, fine-tuned with CTC loss
• Language: English only
• License: Apache 2.0

Usage

crispasr --backend wav2vec2 -m hubert-large-ls960-ft-q4_k.gguf -f audio.wav

Files

File	Size	JFK Result
hubert-large-ls960-ft-f16.gguf	627 MB	perfect
hubert-large-ls960-ft-q4_k.gguf	212 MB	perfect

Provenance and EU AI Act Art. 53 note

• Upstream model: facebook/hubert-large-ls960-ft — published by facebook.
• Upstream licence: apache-2.0. This repository redistributes under the same terms; it grants no rights the upstream licence does not.
• What was done here: format conversion and/or quantisation only (GGUF). No training, no fine-tuning, no merging, no distillation, no change to architecture, vocabulary or capability. Only the numeric representation of the upstream weights differs.
• Training data: documented — where it is documented at all — by the upstream provider; see the upstream model card. No training data was used, added or selected by this repository.
• Provider status: under Regulation (EU) 2024/1689 the upstream authors remain the provider of this model. Converting the serialisation format does not make this repository the provider of a new general-purpose AI model, and no such claim is made. Questions about training content, copyright policy or model capability belong upstream.